🦞 HackMyClaw is live!
$100 to whoever can trick my AI assistant into leaking secrets.env
The rules:
• Email is your only weapon ([email protected])
• Fiu checks mail hourly
• Standard prompt injection won't cut it
Think you can crack an @openclaw agent?
Ricardo Hausmann @ricardo_hausman fundador del @HarvardGrwthLab sobre reformas del gobierno:
“Con eso no van a ir muy lejos. Chile está estancado en ingresos medios, porque le falta el motor fundamental de los países ricos: innovar en nuevos productos y nuevos procesos”
LO QUE IMPORTA: Un país demasiado simple
https://t.co/xGay5113pm
La libre competencia no solo es clave para la eficiencia. Es condición necesaria para la legitimidad del mercado, a defender siempre.
La delación compensada vuelve a probar su valor para detectar carteles.
De probarse, todo el rigor de la ley a los delincuentes coludidos: cárcel.
Many engineering teams are now dealing with this:
- Business stakeholders expect AI to make everything 5x-10x faster.
- For prototypes or a startup environments, this might be true.
- For the median production system (with legacy code, edge cases, security, coordination, and maintenance) the gain is often much smaller. Maybe 5–20%.
This creates tension between speed and stability.
This map is a wild example of how fast babies can fall out of fashion. In Chile fertility rates fell massively between 2018 and 2020. Between 2020 and 2024 rates continued to fall across the nation again! Source: https://t.co/IjsWODRkTn
Reviewing LLM output has become a key part of writing code.
I see a lot of people doing superficial reviews. But I don’t want coding agents to make me think less.
I want them to make me slow down too: show changes piece by piece, explain why each change exists, and explore alternatives before anything gets merged.
I made a small skill for this: diff-by-diff .
It walks through a branch diff one hunk at a time, waits for you to say next, and only commits the current hunk when you say commit.
Link below.